Hand tasks to computer-use agents without the guesswork.
Write a task once as clear steps: what to click, what to type, what success looks like, what to never touch, and when to stop and ask. Then export a tidy runbook for any agent. Everything stays in your browser.
The free editor keeps one runbook in this browser. Download the agent JSON to keep a copy; it can be opened again here.
Task
Use {{VARIABLES}} for values that change each run, and for anything private. Never paste passwords, card numbers or keys into a runbook.
Never do
Stop and ask if
Steps
Preview This is what the agent gets.
How it works
- Steps with checks: each step has an action, an exact target, an optional input, and a success check the agent confirms before moving on.
- Guard rails: "never do" rules, stop conditions, and a "needs approval" flag on any step that's hard to undo.
- Lint: ClickScript flags missing targets and checks, vague words like "somehow" or "etc", risky steps without approval, and anything that looks like a password, key or card number.
- Export: copy the Markdown or plain text into your agent's instructions, or download structured JSON for your own tooling.
ClickScript doesn't run agents or call any AI model. Your runbook is stored only in this browser unless you download it.
Pricing
Free: the full editor, lint, and Markdown, text and JSON export.
Team library (proposed, not final)
Pro (in development): a shared runbook library with versions, and later a browser recorder that fills steps from your clicks. Early access is open now as a $6 one-time pre-order.